Decorrelating jet substructure variables from the jet mass for W boson identification

Non-trivial correlations between substructure variables and the mass of a reconstructed hadronic jet can transform the mass distribution of background jets when such variables are used to identify the ori- gin of the jet. This can affect a physics analysis, e.g. the search for hypothetical heavy resonances at the CMS experiment. Several meth- ods are known today to decorrelate these variables from the mass to some extent. Following a previous study by the ATLAS Collabora- tion, this study aims at testing three such decorrelation methods on two widely used substructure variables. Comparing these methods for both variables, a significant degree of mass decorrelation, as well as an increase in tagging performance, can be achieved by all of these meth- ods, particularly the so-called fixed-efficiency k-nearest-neighbor fit. Their behavior is also shown to be robust with respect to an increased number of primary vertices. Additionally, the results are shown to be comparable to the ones obtained by the ATLAS Collaboration.